On 02/10/2014 04:42 AM, Ben Pfaff wrote: > On Mon, Feb 10, 2014 at 01:42:23AM +0400, Andrey Korolyov wrote: >> On 02/09/2014 11:14 PM, Ben Pfaff wrote: >>> On Sat, Feb 08, 2014 at 05:24:16PM +0400, Andrey Korolyov wrote: >>>> 'ofproto - flow monitoring pause and resume' test failing continuously >>>> during make check on all 2.x major branches. The problem is in >>>> 'ovs-ofctl add-flows br0 flows.txt' execution which actually hangs a >>>> test switch because set count is insanely high and test execution >>>> actually hitting the timeout: >>>> >>>> wc -l ./tests/testsuite.dir/0613/flows.txt >>>> 1400833 ./tests/testsuite.dir/0613/flows.txt >>>> >>>> filled with cookie=1,reg1=[incrementing count],actions=drop >>>> >>>> Can anyone please fix this? >>> >>> We don't see the problem. What's different about your system? >>> >> >> I am observing this behavior on wheezy with a couple of backports and on >> gentoo on my laptop. Generic wheezy passes test just fine, I will try to >> pin the issue and report in case of success. > > What's in /proc/sys/net/core/rmem_max? >
Ben, thanks for this suggestion, you saved me a couple of hours - non-default network memory settings led to this. Okay, and I almost immediately faced another issue - check passes well as long as I am not trying to build datapath at once, with kernel datapath 'ofproto-dpif - NetFlow active expiration - IPv6 collector' failing. Though it is 2.1 still-experimental branch, I am willing to play first with it due to easier mechanism to fake ARP responses. Please check the attached log for details.
failed-dpif-deletion.log.gz
Description: GNU Zip compressed data
_______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
